Abstract
Pairs of Mondrians subtending 5.5 deg were presented in succession, each for 1 sec, in darkness, on a calibrated monitor. The Mondrians simulated 49 different Munsel papers, chosen at random on each trial. Illuminants were unpredictably either (a) 4000 K (first Mondrian) and 6700 K (second), or (b) 16000 K (first) and 6700 K (second). The simulated illuminant on the central patch was offset by one of 9 steps along the daylight locus centred on 6700 K. Five naive observers rated the central patch on a 0% to 100% scale, extending the typical binary choice. A Brunswickian color constancy index (CCI) was estimated from mean ratings. CCI was 18% in (a) and 32% in (b) when rating hue and saturation, and 72% in (a) and 96% in (b) when rating the simulated material (the ‘hue-saturation’ and ‘paper’ matches of Arend & Reeves, JOSA, 1986). Clearly color constancy is task-dependent, even when observers cannot adjust the stimuli (as they had done in Arend & Reeves). CCI was unaffected by narrow black lines interposed between the patches, and by a brief (0.2 sec) dark gap separating the Mondrians, even though the gaps appeared to break up the display. We conclude that if the signals which support color constancy in Mondrians are changes in local edge-ratios at the cone level (Foster & Nasciamento, Proc. R. Soc. B, 1994), these ratios depend on encodings which can span across clearly visible spatial and temporal gaps.
EPSRC